A standing maintenance window is set for 8:00 AM to 5:00 PM on the first Tuesday of every month. This time frame is reserved for changes, upgrades, and enhancements to be made to the system which may be impactful to users and as such cannot be done while Kamiak is online and jobs are running. However, Kamiak will only be taken offline if there is a need to do so. If there are no maintenance tasks to do, Kamiak will remain online through the maintenance window and jobs will continue to be scheduled and run.
When maintenance tasks require Kamiak to be taken offline, an announcement will be made at least 1 week prior to the start of the planned outage. Additionally, a reservation will be placed into Kamiak’s resource scheduler to prevent compute jobs from running during the window. This means that when a maintenance outage is announced, newly submitted jobs which have a wall clock limit that would run into the window will not be scheduled to run. Such jobs will remain queued until the maintenance is complete. Users with very long-running jobs that cannot complete in time may have their jobs canceled and are highly encouraged to integrate checkpointing into their jobs.